[Feature] timestamps for TTS via a forced-aligner pooling (#4795)

Signed-off-by: wjinxu <1299461899@qq.com>
Co-authored-by: xurui <xurui1@modelbest.onaliyun.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

vllm_omni/entrypoints/openai/api_server.py

@@ -1127,7 +1127,11 @@ async def omni_init_app_state(
         state.openai_serving_models,
         request_logger=request_logger,
         model_name=model_name,
-        forced_aligner_config=build_forced_aligner_config(args),
+        forced_aligner_enabled=build_forced_aligner_config(
+            getattr(args, "forced_aligner", None),
+            getattr(args, "forced_aligner_config", None),
+        )
+        is not None,
     )
 
     # Warm up speech pipeline (CUDA Graph capture, torch.compile) so the first